Re: Skybuck presents Faster Div 7

"Robert Redelmeier" <redelm@[EMAIL PROTECTED]
> ha scritto nel messaggio
news:F9lQj.22382$%41.12486@[EMAIL PROTECTED]
> In alt.lang.asm Wolfgang Kern <nowhere@[EMAIL PROTECTED]
> wrote in part:
>> Robert Redelmeier mentioned:
>>> 'scuse me for the interruption, but exactly how could this
>>> iteration be faster than MUL by reciprocals (small LUT) ?
>>
>> Oh yes, I forgot to mention this option ;)
>>
>> MUL by biased 2**n/7
>> ****ft right n
>
>
> Even easier with 7 -- Use 32 for n:
>
>  MOV  edx, 613556756     ; or faster in a LUT
>  MUL  edx                ; dividend in eax
>  XCHG edx, eax           ; result in eax

are sure that have ****ft for n and not for n+2?

the above is all right? give it the correct result?
i prove the result of the below program (assembly + C)
seems all right
----------------
;  for  linux
;     mu.asm
; BL:$ nasm -felf mu.asm
; BL:$ gcc prova.c mu.o -o prova123
; BL:$ gdb ./prova123

section DATA
global div7
section TEXT

; 0ra, 4par */
div7:
mov eax, 0x92492493
mov edx, [esp+4]
mul edx
mov eax, edx
shr eax, 2
ret

-----------------------------------
/*   mu.m
/* BL:$ ./asc mu.m mu.asm
/* BL:$ nasm -felf mu.asm
/* BL:$ gcc prova.c mu.o -o prova123
/* BL:$ gdb ./prova123
section DATA
global div7
section TEXT

/* 0ra, 4par */
div7: a=0x92492493; r=[s+4]; mul r; a=r; a>>=2; ret
--------------------

/*  prova. c   */
#include <stdio.h>

unsigned div7(unsigned);

int main(void)
{unsigned a, b;
  printf("Inserisci un numero> "); fflush(stdout);
  if( scanf("%u", &a)!= 1)
              { printf("Error\n"); return 0;}
  b=div7(a);
  printf("%u/%u=%u\n", a, 7, b);
  return 0;
}

the reference for the way of compute these numbers for "div"
are by Terje Mathisen.

> AFAICS this doesn't require any rounding clean-up.
>
> -- Robert